Developing epidemic forecasting models to assist disease surveillance for influenza with electronic health records

Seasonal influenza is an important infectious disease monitored by the Centers for Disease Control. Monitoring, early detection, and prediction of influenza outbreaks can accelerate life-saving public health responses. However, ensuring completeness and timeliness of all surveillance systems can be difficult. Current influenza surveillance systems can be supported through the estimation of influenza activity, which we aim to achieve by defining influenza cases through electronic medical records (EMRs) from Chang Gung Memorial Hospital and constructing predictive models for predicting the epidemic trend for influenza. In total, 84,820 influenza cases were matching the case definition in 2010–2015. Using the exponential smoothing model, autoregressive integrated moving average (ARIMA) model, and Prophet model, we trained the models with influenza cases from 2010 to 2014; we then renewed the models at different intervals and compared their capacities. The results show that the ARIMA model performs best in both Linkou and Kaohsiung branches. In conclusion, we successfully constructed an influenza epidemic predictive model based on EMRs, which may be used as a decision support tool for improving influenza surveillance, controlling infection, and managing resources.